Questions about preventing AI cheating

Aiseptor is a network-layer security platform that assessment providers embed to prevent AI-assisted cheating on any candidate device. It enforces a per-session, default-deny policy at the operating-system and network layer, so invisible AI overlays, on-device LLMs, remote-access tools, and second-device pivots cannot reach what they depend on. Aiseptor deploys in about 30 seconds, uses no webcam or kernel driver, and removes itself when the session ends.

Network-layer exam security enforces integrity at the network path a candidate's device travels, rather than at the browser window or through a webcam. A per-session policy allows only the destinations an exam explicitly needs and blocks everything else — including AI inference endpoints, remote-access relays, and model-hosting services — at the OS and network layer. It governs the device regardless of which application makes a request, so it catches threats that application-layer tools (lockdown browsers) and screen-layer tools (camera proctoring) cannot see.

A lockdown browser controls a single exam window at the application layer — tabs, copy-paste, navigation. Anything running outside that window (an invisible overlay, an on-device LLM, a second device, a remote-access helper) is invisible to it, and public bypasses exist. Aiseptor enforces beneath the browser, at the OS and network layer, so it governs the whole device. Aiseptor also now offers Aiseptor Secure Browser (in beta), a lockdown browser for Mac and Windows backed by that same enforcement — so one platform can cover the application, device, and network layers.

No. Those providers operate the physical and behavioral layer — webcam, gaze tracking, room scans, identity. Aiseptor operates the device and network layer beneath them. They are complementary: a proctoring provider can embed Aiseptor to solve the device-side AI-cheating problem their in-browser and camera tooling cannot reach, and high-stakes programs typically pair the two.

Yes. Aiseptor blocks Cluely-class invisible overlays by the technique they all share, not by a list of tool names — so renamed and never-before-seen overlays are caught the same way the known ones are. It detects the screen-capture-exclusion flag every hidden overlay must set, the GPU memory their inference requires, and active-overlay window signals, and it blocks the network path to the AI endpoints they depend on with a default-deny policy. An overlay that cannot reach a model is inert, whatever it is named.

Yes. A locally run model produces no external API traffic, so network blocking alone will not catch it. Aiseptor detects on-device LLMs through OS-level signals — active local-inference processes, model-hosting services, and the GPU/memory footprint inference requires — and blocks the model-hosting and update endpoints those tools reach at the network layer. This is a threat that webcam proctoring and lockdown browsers cannot see at all.

Aiseptor does not rely on process-name denylists, which go blind the moment a tool is renamed or recompiled. It detects the invariant techniques shared across tools — the screen-capture-exclusion flag, inference GPU/memory signatures, and the outbound network path to AI services — so renamed, forked, and not-yet-released tools trip the same controls as known ones.

Yes. Remote-access and screen-share tools (AnyDesk, TeamViewer, Chrome Remote Desktop, conferencing screen-shares) are blocked at the session boundary, so an off-screen helper cannot reach the device and a local agent cannot phone home. Because the policy applies across all network interfaces, a second device or mobile hotspot used to tunnel AI traffic is covered rather than becoming an unmonitored path.

No. Aiseptor uses no webcam, no microphone, no keystroke logging, and no screen recording. It enforces integrity by controlling what the device can reach, not by recording the candidate. This removes the privacy exposure, bias risk, and dispute surface of camera-based proctoring while still blocking AI overlays, on-device LLMs, and remote-access tools.

Aiseptor collects network-access requests made during the session (to enforce the approved-domain policy), device-activity signals observed during the session window, and signed session metadata for the audit trail. It does not collect webcam, microphone, keystroke, or screen-recording data. Default data retention is 24 hours. Aiseptor is GDPR and CCPA compliant, with SOC 2 and ISO 27001 in progress.

Yes. Aiseptor is used by enterprise hiring teams to prevent AI cheating in technical and coding interviews on platforms like Codility, HackerRank, and CodeSignal. It blocks invisible coding-assistant overlays, on-device LLMs, and remote-access helpers by closing the network path they need — only the coding platform and approved documentation remain reachable — so candidates are evaluated on their own ability, without a webcam.

Yes. Aiseptor secures remote certification exams without test centers by deploying a per-session security enclave on the candidate's own device, enforcing a default-deny network policy, and producing a cryptographically signed audit trail. This prevents proxy test-takers, AI overlays, and remote-control assistance at the OS and network layer, giving certification bodies test-center-grade assurance — and a tamper-evident record per session — without the cost or geographic limits of physical centers.

Yes. Aiseptor integrates through a plug-and-play REST API and is designed to drop into an existing assessment, LMS, or ATS flow with no UI changes — your platform receives a signed webhook at session end with the integrity verdict and a link to the audit report. Integration typically takes under one business day.

Yes. Aiseptor Secure Browser (in beta) is a lockdown / secure exam browser that runs natively on both macOS and Windows with parity, so Mac candidates get the same enforcement as Windows candidates rather than a degraded or unsupported experience. Unlike browser-only lockdown tools, it is backed by Aiseptor's OS- and network-layer enforcement, so it also blocks AI overlays, on-device LLMs, and remote-access tools that run outside the browser.

A normal lockdown browser operates only at the application layer and is blind to anything outside the exam window. Aiseptor Secure Browser provides the same window lockdown but is backed by a default-deny network policy and OS-level signal detection, so it governs the actual path between the device and any external service. It also requires no webcam and no kernel driver, and removes itself when the session ends.

Aiseptor is usage-based: you pay per exam session, with every defense layer included and no commitment. New users can start with free sessions. Public rates are being refreshed — request access for current per-session pricing, or contact us about volume.

Aiseptor deploys on a candidate's device in about 30 seconds in user space — no kernel driver, no admin rights, no persistent install — and platform integration typically takes under one business day. You can start with free sessions before any commitment.

Aiseptor was founded in 2025 by Akshay Aggarwal, Divya Bhanushali, and Sanjay Ram. It won the 2026 Cornell Tech Startup Award for its network-layer approach to preventing AI cheating, and is an NVIDIA Inception member. In pilots, Aiseptor recorded a 0% bypass rate and zero false positives across 25+ attempted attack vectors.